I’ve spent the past few years thinking about how to create legible ML models, and the past few months applying those ideas. I now have a library of public-domain scripts that let me feature-select, constrain, and fit a model which can then be described using simple graphs while maintaining similar performance to XGBoost and other ‘standard’ ML algorithms. I want to try my approach in the real world, and finding non-toy Data Science problems that people want help with has been surprisingly hard thus far.
I’d describe my approach and motivation in more detail, but akur8 has me covered: I do what they do, except less professionally, less expensively, more flexibly, without proprietary code, and without the exclusive focus on insurance pricing (though my approach is optimized for that, since it’s where I started). Also, I have a truly marvellous method for modelling interactions between features which this margin is too small to contain (tl;dr fit multiple models in parallel).
I’d also be (less) interested in any kind of Analysis work. If anyone wants to hand me a dataset and some vague instructions, then get back a report that looks kind of like this (but better, since I have the benefit of three years’ extra experience), I’d be up for that.
I’m willing to work for free on EA tasks, and would ask for only a nominal fee to work on anything else; I may end up raising my prices later, but right now I just want to apply my skills to something meaningful ASAP. If your organisation is such that recruiting a random stranger you met on an internet forum to do Data Science/Data Analysis seems both tempting and viable, I want to talk to you.
Any takers? Or any helpful opinions on what I’m doing?
ETA: This post somehow led to me getting a full-time job; I’m no longer actively seeking out Data work at time of writing; I have updated significantly in the direction of “asking for things on LessWrong is a good idea”.
[Question] Does anyone have any Data Sidequests?
I’ve spent the past few years thinking about how to create legible ML models, and the past few months applying those ideas. I now have a library of public-domain scripts that let me feature-select, constrain, and fit a model which can then be described using simple graphs while maintaining similar performance to XGBoost and other ‘standard’ ML algorithms. I want to try my approach in the real world, and finding non-toy Data Science problems that people want help with has been surprisingly hard thus far.
I’d describe my approach and motivation in more detail, but akur8 has me covered: I do what they do, except less professionally, less expensively, more flexibly, without proprietary code, and without the exclusive focus on insurance pricing (though my approach is optimized for that, since it’s where I started). Also, I have a truly marvellous method for modelling interactions between features which this margin is too small to contain (tl;dr fit multiple models in parallel).
I’d also be (less) interested in any kind of Analysis work. If anyone wants to hand me a dataset and some vague instructions, then get back a report that looks kind of like this (but better, since I have the benefit of three years’ extra experience), I’d be up for that.
I’m willing to work for free on EA tasks, and would ask for only a nominal fee to work on anything else; I may end up raising my prices later, but right now I just want to apply my skills to something meaningful ASAP. If your organisation is such that recruiting a random stranger you met on an internet forum to do Data Science/Data Analysis seems both tempting and viable, I want to talk to you.
Any takers? Or any helpful opinions on what I’m doing?
ETA: This post somehow led to me getting a full-time job; I’m no longer actively seeking out Data work at time of writing; I have updated significantly in the direction of “asking for things on LessWrong is a good idea”.
ETA^2: I’m now actively looking for work again.